DeLinker

DeLinker generates 3D-aware molecular linkers and molecules using a graph-based deep generative model to support fragment linking, scaffold hopping, and PROTAC design.

Key Features:

• Graph-based deep generative model: DeLinker employs a graph-based deep generative model that integrates machine learning with structural molecular representations.
• 3D structural integration: The model directly incorporates three-dimensional structural information into the generative process.
• Protein-context-dependent generation: Generation conditions on the relative distance and orientation between two input fragments or partial structures to produce context-compatible linkers.
• Two operational settings: Supports a fixed atom number setting that matches a reference molecule and a specified atom number setting that generates linkers with a predetermined atom count.
• Performance vs database baselines: In large-scale evaluations it produced 60% more molecules with high 3D similarity overall and outperformed baselines by 200% for linkers containing at least five atoms.

Scientific Applications:

• Fragment Linking: Combines two molecular fragments into cohesive structures while maintaining high three-dimensional similarity to the originals.
• Scaffold Hopping: Explores novel chemical scaffolds by generating linkers that connect different molecular cores to enable scaffold replacement.
• PROTAC Design: Assists in designing proteolysis targeting chimeras (PROTACs), which are bifunctional molecules that target specific proteins for degradation.

Methodology:

Graph-based deep generative modeling that integrates three-dimensional structural information and conditions on the relative distance and orientation between two input fragments or partial structures; provides fixed atom number and specified atom number linker generation settings.